Study Of Resource Allocation Optimizationin For Multi-cell Network Based On Deep Q Network

With the rapid development of communication networks,mobile users have an increasingly high demand for data,which promotes the construction of cellular network infrastructure and the explosive growth of mobile devices.However,too dense base station deployment affects the communication service quality between mobile users and devices,and the consumption of wireless resources also increase exponentially.Therefore,how to optimize the allocation of wireless resources become the key to research.In the field of communication,the optimization of resource allocation requires dealing with a large amount of base station data.With the increasing prosperity of artificial intelligence,scholars combine it with data processing at the base station to optimize related processing and results.As the core technology of artificial intelligence research,machine learning algorithms increases the diversity of data processing methods,reduces the high computational consumption involved in the processing of traditional technologies and also improves algorithmic decision-making capabilities.Reinforcement Learning(RL)is one of the main methods of machine learning.It is introduced into wireless communication systems.Its ability to simulate the network environment and optimize the allocation of environmental resources is particularly prominent.But reinforcement learning can only deal with static network environments,and the algorithm complexity is very high.Therefore,this thesis introduces Convolutional Neural Networks(CNN)and proposes Deep Q Network(DQN)to construct a dynamic continuous state space to obtain better system performance.This thesis will propose wireless resource allocation optimization of spectrum efficiency(SE)and energy efficiency(EE)based on deep reinforcement learning algorithm for the downlink of a multi-cell cellular network,and complete the following works:Firstly,a multi-cell spectrum efficiency optimization algorithm based on deep reinforcement learning is proposed.The problem of cell network system capacity is modeled.Q learning algorithm is used to maximize the capacity of the whole network.Considering the huge base station data and the constantly changing network environment,a wireless resource mapping method based on deep reinforcement learning algorithm is proposed.Through simulation analysis,compared with the traditional Q-learning,the proposed deep reinforcement learning algorithm can obtain higher system capacity and significantly improve the convergence and stability.Secondly,a resource allocation algorithm based on Dueling DQN for joint optimization of spectrum efficiency and energy efficiency is proposed.According to the system model,the energy efficiency and spectrum efficiency of the base station are obtained respectively.Taking into account the dynamic nature of the environment,SE and EE are dynamically weighted according to the changes in the number of users.Then,the CNN in the DQN is optimized.The Dueling DQN algorithm is proposed to improve the stability of the algorithm.The simulation results show that,compared with the DQN algorithm,the Dueling DQN algorithm better solves the combined optimization problem of SE and EE,and the convergence of the algorithm is also improved.